On Thu, 2005-06-30 at 17:00 -0600, Brett Glass wrote:
> At 04:48 PM 6/30/2005, Henrik Nordstrom wrote:
>
> >It is planned to implement a real RAM based cache in Squid, it just isn't
> >really there yet.
>
> It would help as we move toward peak traffic levels of 500 to 1000 hits per
> second. At
> that level, even a cache with multiple drives becomes noticeably disk-bound.
> Right
> now, adding RAM doesn't help, because popular objects don't wind up there
> unless they're
> new. We were baffled as to why the extra RAM wasn't helping until we did some
> tests and
> saw that the TCP_HITs weren't followed by TCP_MEM_HITs on the same URLs.
>
> We could create a "memory disk," but populating it is awkward.
Huh? Just put your cache_dir in a ramdisk or tmpfs (i.e. /dev/shm on
recent Linuxes, /tmp on Solaris, don't really know about other OSes..)
Kinkie
